DEV Community

Cover image for Hello Again Attack: Windows Hello & BitLocker PIN Risk
devTalk
devTalk

Posted on Originally published at dev-talk.com

Hello Again Attack: Windows Hello & BitLocker PIN Risk

Last updated: 18 November 2026. Based on the research paper (arXiv 2611.18126, ACM CCS '26). Vendor fixes were still pending when I checked. SeeHow this was checked.

A short PIN is safe on a Windows PC for one reason: the TPM stops answering after a few wrong guesses. Researchers at TU Berlin showed that on many machines this lockout can be reset by rolling back the TPM's stored state, so the attacker never reaches the limit. The attack needs physical access and soldering-level hardware work, so it is not a drive-by exploit. It does change how much a four-digit PIN is worth on a stolen laptop, and that is what this article is about.

The short version

  • What breaks: the TPM's failed-guess counter (dictionary attack protection) can be reset on affected systems, so PIN guessing continues past the lockout.
  • What is affected: Windows Hello PINs and BitLocker TPM+PIN protectors on systems where the firmware TPM keeps its state in external SPI flash.
  • What the attacker needs: the device in hand, plus hardware work on the flash chip (one path also needs voltage glitching; another needs only a flash programmer).
  • What you can do now: longer PINs, enhanced (alphanumeric) BitLocker startup PINs, and keeping firmware current. Details below.

How the lockout normally works

Windows Hello PINs and BitLocker TPM+PIN rely on the TPM's dictionary attack protection. After a number of failures the TPM refuses further guesses for a while. The paper uses a default rate of one try per 10 minutes as its baseline for a naive attacker.

On many x86 systems the TPM is a firmware TPM (fTPM) running inside the CPU's security processor. It has no storage of its own, so its non-volatile state, including the lockout counters, is kept on the motherboard's SPI flash chip. That creates an obvious weakness: whoever controls the flash can restore an older copy of it, and the counter goes back with it. To defend against this, some flash chips implement RPMC, Replay Protected Monotonic Counters (JEDEC JESD260), which are counters that can only go up and are authenticated with an HMAC key.

What the paper found

The authors (Werling, Finder, Jacob, Prein, Kühnapfel and Seifert, TU Berlin) found ways around this protection on several platforms:

Weakness Where Effect
HMAC key leak by voltage glitching Winbond RPMC flash The chip returns its HMAC key, which lets the attacker forge counter responses. The paper measured about 1 second per leak.
Unverified counter increment AMD controller The controller does not check that an increment succeeded. An interposer board can suppress the increment command and reply with a fake success.
No RPMC flash at all Two AMD desktops tested (Zen 2 and Zen 5) Re-flashing an old SPI image with a programmer reset the lockout counters. No special interposer was needed.
Constant key data Intel RPMC controllers Classed by the paper as a weakness, not a vulnerability. It lowers the cost of exploiting the Winbond key leak.
Counter overflow, corruption Winbond, Macronix, GigaDevice flash Counters can overflow at their maximum (roughly 4 to 6 weeks of incrementing in the authors' tests), and GigaDevice counters could be corrupted by glitching (about 10% success per attempt).

What an attacker actually needs

This is why the risk is real but narrow. From the paper:

  • Physical possession of the machine.
  • For the interposer paths, desoldering the flash chip and fitting a socket, plus an interposer built from inexpensive parts (the authors used a Raspberry Pi Pico 2W, roughly $5, with a multiplexer and level shifting).
  • For the Winbond key leak, voltage glitching equipment.
  • Automation to type PINs and reboot repeatedly.
  • No knowledge of the PIN. Note also that Windows itself limited the Windows Hello demo to 5 tries per boot.

Someone who steals a laptop at an airport is unlikely to do this on the spot. A motivated attacker with a lab and a target worth the effort, or someone who has your laptop overnight, is the realistic threat model.

How fast it was in practice

Measured results from the paper (236 PIN attempts in each run):

  • Windows Hello on an AMD laptop: 57 minutes net, about 4.1 tries per minute.
  • BitLocker TPM+PIN on an Intel laptop (Framework Laptop 13): 8 min 40 s in the first run and 6 min 55 s after the automation was optimized, about 34 tries per minute.

The worst-case times for exhausting the PIN space are estimates derived from those rates:

PIN With lockout (naive, 10 min per try) With counter rollback
4-digit Windows Hello PIN 69 days 40 hours
6-digit Windows Hello PIN not listed here 168 days
6-digit BitLocker PIN 19 years 20 days

The paper also estimates that moving Hello from 4 to 6 digits and BitLocker from 6 to 8 would raise the rollback worst case to 168 days (Hello) and about 6 years (BitLocker). These are worst cases for numeric PINs.

Who has said what

  • Microsoft: according to the paper, closed the Windows Hello report as moderate severity (defense in depth), with no CVE and no bounty, and committed to raising the recommended minimum Hello PIN length to 6 digits. The paper says Microsoft will communicate when its fix ships.
  • AMD: acknowledged the report and passed it on for internal review. As of the paper, no committed fix. Some news coverage of the paper cites an AMD bulletin and CVE for this; I could not open AMD's site from my environment, and the paper itself lists no CVE, so I am not quoting one. Check AMD's product security page directly.
  • Intel: first closed the key-reuse report as informative, then classified it as a security weakness (not a vulnerability) after the case was reopened.
  • Winbond: evaluating; a silicon revision is one option. No committed fix or date. Flash hardware already deployed cannot be patched by software alone.
  • Macronix: confirmed the counter wraparound and closed it as impractical.
  • GigaDevice: engaged; fix or statement pending.

What to do about it

For individuals

  • Use a longer PIN. Each extra digit multiplies the search space by ten, which is the only lever you control.
  • For a laptop that leaves the house, use a BitLocker startup PIN and not just the TPM, and prefer an alphanumeric one (below).
  • Keep UEFI and firmware updates current. AMD and the flash vendors have not shipped fixes for everything yet, but updates are where they will arrive.
  • Treat a laptop that was out of your sight for hours, with a short PIN, as potentially accessed.

For administrators

BitLocker's startup PIN length and character set are controlled by two Group Policy settings under Computer Configuration > Administrative Templates > Windows Components > BitLocker Drive Encryption > Operating System Drives:

  • Configure minimum PIN length for startup: the documented range is 4 to 20 characters. Raise the minimum.
  • Allow enhanced PINs for startup: when enabled, new startup PINs can include letters, symbols and spaces, not only digits. Not all hardware supports enhanced PINs in the pre-boot environment, so test on your models before rolling it out.

Existing protectors do not change when you change the policy. Users must set a new PIN. I did not run the following on Windows (this environment is Linux), but the standard way to inspect and replace the protector from an elevated prompt is:

manage-bde -status C:
manage-bde -protectors -delete C: -type TPMAndPIN
manage-bde -protectors -add C: -TPMAndPIN
Enter fullscreen mode Exit fullscreen mode

Make sure you have a recovery key backed up before deleting protectors, and run the commands from an elevated session. Check manage-bde -? on your Windows build if the syntax differs.

Checking your AMD machine on Linux

The paper notes that on AMD systems, Linux 5.19 and later expose rpmc_spirom_available and rpmc_production_enabled through sysfs, and that fwupdmgr security reports the status of AMD SPI replay protection. I did not run these here because the sandbox has no AMD hardware, so I cannot show output. They tell you whether your board has RPMC support, which is a precondition, not proof that you are safe.

Common misreadings

  • "Windows Hello is broken for everyone." The demonstrated attacks need physical access and hardware modification. Remote attackers are not in scope.
  • "A TPM is useless." The TPM still protects against online and software guessing. The issue is the storage the firmware TPM depends on.
  • "Dedicated TPM chips are affected the same way." The paper's focus is firmware TPMs storing state on SPI flash. I did not verify the effect on discrete TPMs.

Frequently asked questions

Is there a CVE for the Hello Again attack?

The paper lists none, and says Microsoft issued no CVE for the Windows Hello case. Some secondary reports mention an AMD identifier; I could not confirm it from an AMD source.

Can someone do this remotely?

Not according to the paper. Every demonstrated path involves physical access to the flash chip.

Is a six-digit PIN enough?

It is better, not safe. The paper's estimate for a 6-digit Windows Hello PIN with rollback is 168 days worst case, and for a 6-digit BitLocker PIN it is 20 days. For high-value laptops, use a longer or alphanumeric BitLocker PIN.

Does a Microsoft update fix it?

Not yet, as far as I could find. Microsoft committed to a longer recommended Hello PIN, and fixes for the hardware flaws depend on AMD, the flash vendors and BIOS updates.

Does BitLocker without a PIN get worse?

This research targets PIN guessing. TPM-only BitLocker has a different exposure that I did not cover here.

update log

  • 18 November 2026: first version, days after the paper appeared.
  • To recheck: Microsoft's final guidance on Hello PIN length, AMD's firmware and bulletin status, Winbond and GigaDevice statements, and any CVE assigned.

Sources


Originally published on DEV Talk.

Top comments (0)